5 # bug number: 13129 13129 10124
6 ALWAYS_EXCEPT="2 3 15c $REPLAY_DUAL_EXCEPT"
9 PTLDEBUG=${PTLDEBUG:--1}
10 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
12 CLEANUP=${CLEANUP:-""}
13 MOUNT_2=${MOUNT_2:-"yes"}
14 . $LUSTRE/tests/test-framework.sh
18 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
21 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 5 14"
25 cleanup_and_setup_lustre
27 rm -rf $DIR/[df][0-9]*
29 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
33 replay_barrier $SINGLEMDS
37 checkstat $MOUNT2/a || return 1
38 checkstat $MOUNT1/b || return 2
39 rm $MOUNT2/a $MOUNT1/b
40 checkstat $MOUNT1/a && return 3
41 checkstat $MOUNT2/b && return 4
45 run_test 1 "|X| simple create"
49 replay_barrier $SINGLEMDS
53 checkstat $MOUNT2/adir || return 1
55 checkstat $MOUNT2/adir && return 2
58 run_test 2 "|X| mkdir adir"
61 replay_barrier $SINGLEMDS
63 mkdir $MOUNT2/adir/bdir
66 checkstat $MOUNT2/adir || return 1
67 checkstat $MOUNT1/adir/bdir || return 2
68 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
69 checkstat $MOUNT1/adir && return 3
70 checkstat $MOUNT2/adir/bdir && return 4
73 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
77 replay_barrier $SINGLEMDS
78 mkdir $MOUNT1/adir && return 1
79 mkdir $MOUNT2/adir/bdir
82 checkstat $MOUNT2/adir || return 2
83 checkstat $MOUNT1/adir/bdir || return 3
85 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
86 checkstat $MOUNT1/adir && return 4
87 checkstat $MOUNT2/adir/bdir && return 5
90 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
94 # multiclient version of replay_single.sh/test_8
96 multiop_bg_pause $MOUNT2/a o_tSc || return 1
99 replay_barrier $SINGLEMDS
101 wait $pid || return 1
104 [ -e $MOUNT2/a ] && return 2
107 run_test 5 "open, unlink |X| close"
112 multiop_bg_pause $MOUNT2/a o_c || return 1
114 multiop_bg_pause $MOUNT1/a o_c || return 1
117 replay_barrier $SINGLEMDS
119 wait $pid1 || return 1
123 wait $pid2 || return 1
124 [ -e $MOUNT2/a ] && return 2
127 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
130 replay_barrier $SINGLEMDS
131 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
133 checkstat $MOUNT2/$tfile || return 2
134 rm $MOUNT1/$tfile || return 3
138 run_test 8 "replay of resent request"
141 replay_barrier $SINGLEMDS
142 mcreate $MOUNT1/$tfile-1
143 mcreate $MOUNT2/$tfile-2
144 # drop first reint reply
145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
147 do_facet $SINGLEMDS lctl set_param fail_loc=0
149 rm $MOUNT1/$tfile-[1,2] || return 1
153 run_test 9 "resending a replayed create"
156 mcreate $MOUNT1/$tfile-1
157 replay_barrier $SINGLEMDS
158 munlink $MOUNT1/$tfile-1
159 mcreate $MOUNT2/$tfile-2
160 # drop first reint reply
161 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
163 do_facet $SINGLEMDS lctl set_param fail_loc=0
165 checkstat $MOUNT1/$tfile-1 && return 1
166 checkstat $MOUNT1/$tfile-2 || return 2
171 run_test 10 "resending a replayed unlink"
174 replay_barrier $SINGLEMDS
175 mcreate $MOUNT1/$tfile-1
176 mcreate $MOUNT2/$tfile-2
177 mcreate $MOUNT1/$tfile-3
178 mcreate $MOUNT2/$tfile-4
179 mcreate $MOUNT1/$tfile-5
180 # drop all reint replies for a while
181 do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
182 # note that with this fail_loc set, facet_failover df will fail
183 facet_failover $SINGLEMDS
184 #sleep for while, let both clients reconnect and timeout
185 sleep $((TIMEOUT * 2))
186 do_facet $SINGLEMDS lctl set_param fail_loc=0
188 rm $MOUNT1/$tfile-[1-5] || return 1
192 run_test 11 "both clients timeout during replay"
195 replay_barrier $SINGLEMDS
197 multiop_bg_pause $DIR/$tfile mo_c || return 1
200 #define OBD_FAIL_LDLM_ENQUEUE 0x302
201 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
202 facet_failover $SINGLEMDS
203 do_facet $SINGLEMDS lctl set_param fail_loc=0
204 df $MOUNT || return 1
207 kill -USR1 $MULTIPID || return 3
208 wait $MULTIPID || return 4
209 $CHECKSTAT -t file $DIR/$tfile || return 2
214 run_test 12 "open resend timeout"
217 multiop_bg_pause $DIR/$tfile mo_c || return 1
220 replay_barrier $SINGLEMDS
222 kill -USR1 $MULTIPID || return 3
223 wait $MULTIPID || return 4
226 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
227 facet_failover $SINGLEMDS
228 do_facet $SINGLEMDS lctl set_param fail_loc=0
229 df $MOUNT || return 1
232 $CHECKSTAT -t file $DIR/$tfile || return 2
237 run_test 13 "close resend timeout"
240 replay_barrier $SINGLEMDS
241 createmany -o $MOUNT1/$tfile- 25
242 createmany -o $MOUNT2/$tfile-2- 1
243 createmany -o $MOUNT1/$tfile-3- 25
246 facet_failover $SINGLEMDS
247 # expect failover to fail due to missing client 2
248 df $MOUNT && return 1
251 # first 25 files should have been replayed
252 unlinkmany $MOUNT1/$tfile- 25 || return 2
254 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
257 run_test 14 "timeouts waiting for lost client during replay"
259 test_15a() { # was test_15
260 replay_barrier $SINGLEMDS
261 createmany -o $MOUNT1/$tfile- 25
262 createmany -o $MOUNT2/$tfile-2- 1
265 facet_failover $SINGLEMDS
266 df $MOUNT || return 1
268 unlinkmany $MOUNT1/$tfile- 25 || return 2
269 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
271 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
274 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
277 replay_barrier $SINGLEMDS
278 for ((i = 0; i < 2000; i++)); do
279 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
283 facet_failover $SINGLEMDS
285 df $MOUNT || return 1
287 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
290 run_test 15c "remove multiple OST orphans"
293 replay_barrier $SINGLEMDS
294 createmany -o $MOUNT1/$tfile- 25
295 createmany -o $MOUNT2/$tfile-2- 1
298 facet_failover $SINGLEMDS
300 facet_failover $SINGLEMDS
301 df $MOUNT || return 1
303 unlinkmany $MOUNT1/$tfile- 25 || return 2
305 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
309 run_test 16 "fail MDS during recovery (3571)"
312 createmany -o $MOUNT1/$tfile- 25
313 createmany -o $MOUNT2/$tfile-2- 1
315 # Make sure the disconnect is lost
322 df $MOUNT || return 1
324 unlinkmany $MOUNT1/$tfile- 25 || return 2
326 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
330 run_test 17 "fail OST during recovery (3571)"
332 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
335 test_18() { # bug 3822 - evicting client with enqueued lock
337 mkdir -p $MOUNT1/$tdir
338 touch $MOUNT1/$tdir/f0
339 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
340 statmany -s $MOUNT1/$tdir/f 1 500 &
343 do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b # hold enqueue
345 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
346 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
348 usleep 500 # wait to ensure first client is one that will be evicted
349 openfile -f O_RDONLY $MOUNT2/$tdir/f0
351 dmesg | grep "entering recovery in server" && \
352 error "client not evicted" || true
353 do_facet client "lctl set_param fail_loc=0"
354 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
356 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
358 test_19() { # Bug 10991 - resend of open request does not fail assertion.
359 replay_barrier $SINGLEMDS
360 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
362 checkstat $DIR2/${tfile}0 || return 2
363 rm $DIR/${tfile}0 || return 3
367 run_test 19 "resend of open request"
369 equals_msg `basename $0`: test complete, cleaning up
370 SLEEP=$((`date +%s` - $NOW))
371 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
372 check_and_cleanup_lustre
373 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true